2 * Copyright (C) 2001, 2002 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include "dm-uevent.h"
12 #include <linux/init.h>
13 #include <linux/module.h>
14 #include <linux/mutex.h>
15 #include <linux/sched/signal.h>
16 #include <linux/blkpg.h>
17 #include <linux/bio.h>
18 #include <linux/mempool.h>
19 #include <linux/dax.h>
20 #include <linux/slab.h>
21 #include <linux/idr.h>
22 #include <linux/uio.h>
23 #include <linux/hdreg.h>
24 #include <linux/delay.h>
25 #include <linux/wait.h>
27 #include <linux/refcount.h>
29 #define DM_MSG_PREFIX "core"
32 * Cookies are numeric values sent with CHANGE and REMOVE
33 * uevents while resuming, removing or renaming the device.
35 #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
36 #define DM_COOKIE_LENGTH 24
38 static const char *_name
= DM_NAME
;
40 static unsigned int major
= 0;
41 static unsigned int _major
= 0;
43 static DEFINE_IDR(_minor_idr
);
45 static DEFINE_SPINLOCK(_minor_lock
);
47 static void do_deferred_remove(struct work_struct
*w
);
49 static DECLARE_WORK(deferred_remove_work
, do_deferred_remove
);
51 static struct workqueue_struct
*deferred_remove_workqueue
;
53 atomic_t dm_global_event_nr
= ATOMIC_INIT(0);
54 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq
);
56 void dm_issue_global_event(void)
58 atomic_inc(&dm_global_event_nr
);
59 wake_up(&dm_global_eventq
);
63 * One of these is allocated per bio.
66 struct mapped_device
*md
;
70 unsigned long start_time
;
71 spinlock_t endio_lock
;
72 struct dm_stats_aux stats_aux
;
75 #define MINOR_ALLOCED ((void *)-1)
78 * Bits for the md->flags field.
80 #define DMF_BLOCK_IO_FOR_SUSPEND 0
81 #define DMF_SUSPENDED 1
84 #define DMF_DELETING 4
85 #define DMF_NOFLUSH_SUSPENDING 5
86 #define DMF_DEFERRED_REMOVE 6
87 #define DMF_SUSPENDED_INTERNALLY 7
89 #define DM_NUMA_NODE NUMA_NO_NODE
90 static int dm_numa_node
= DM_NUMA_NODE
;
93 * For mempools pre-allocation at the table loading time.
95 struct dm_md_mempools
{
100 struct table_device
{
101 struct list_head list
;
103 struct dm_dev dm_dev
;
106 static struct kmem_cache
*_io_cache
;
107 static struct kmem_cache
*_rq_tio_cache
;
108 static struct kmem_cache
*_rq_cache
;
111 * Bio-based DM's mempools' reserved IOs set by the user.
113 #define RESERVED_BIO_BASED_IOS 16
114 static unsigned reserved_bio_based_ios
= RESERVED_BIO_BASED_IOS
;
116 static int __dm_get_module_param_int(int *module_param
, int min
, int max
)
118 int param
= READ_ONCE(*module_param
);
119 int modified_param
= 0;
120 bool modified
= true;
123 modified_param
= min
;
124 else if (param
> max
)
125 modified_param
= max
;
130 (void)cmpxchg(module_param
, param
, modified_param
);
131 param
= modified_param
;
137 unsigned __dm_get_module_param(unsigned *module_param
,
138 unsigned def
, unsigned max
)
140 unsigned param
= READ_ONCE(*module_param
);
141 unsigned modified_param
= 0;
144 modified_param
= def
;
145 else if (param
> max
)
146 modified_param
= max
;
148 if (modified_param
) {
149 (void)cmpxchg(module_param
, param
, modified_param
);
150 param
= modified_param
;
156 unsigned dm_get_reserved_bio_based_ios(void)
158 return __dm_get_module_param(&reserved_bio_based_ios
,
159 RESERVED_BIO_BASED_IOS
, DM_RESERVED_MAX_IOS
);
161 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios
);
163 static unsigned dm_get_numa_node(void)
165 return __dm_get_module_param_int(&dm_numa_node
,
166 DM_NUMA_NODE
, num_online_nodes() - 1);
169 static int __init
local_init(void)
173 /* allocate a slab for the dm_ios */
174 _io_cache
= KMEM_CACHE(dm_io
, 0);
178 _rq_tio_cache
= KMEM_CACHE(dm_rq_target_io
, 0);
180 goto out_free_io_cache
;
182 _rq_cache
= kmem_cache_create("dm_old_clone_request", sizeof(struct request
),
183 __alignof__(struct request
), 0, NULL
);
185 goto out_free_rq_tio_cache
;
187 r
= dm_uevent_init();
189 goto out_free_rq_cache
;
191 deferred_remove_workqueue
= alloc_workqueue("kdmremove", WQ_UNBOUND
, 1);
192 if (!deferred_remove_workqueue
) {
194 goto out_uevent_exit
;
198 r
= register_blkdev(_major
, _name
);
200 goto out_free_workqueue
;
208 destroy_workqueue(deferred_remove_workqueue
);
212 kmem_cache_destroy(_rq_cache
);
213 out_free_rq_tio_cache
:
214 kmem_cache_destroy(_rq_tio_cache
);
216 kmem_cache_destroy(_io_cache
);
221 static void local_exit(void)
223 flush_scheduled_work();
224 destroy_workqueue(deferred_remove_workqueue
);
226 kmem_cache_destroy(_rq_cache
);
227 kmem_cache_destroy(_rq_tio_cache
);
228 kmem_cache_destroy(_io_cache
);
229 unregister_blkdev(_major
, _name
);
234 DMINFO("cleaned up");
237 static int (*_inits
[])(void) __initdata
= {
248 static void (*_exits
[])(void) = {
259 static int __init
dm_init(void)
261 const int count
= ARRAY_SIZE(_inits
);
265 for (i
= 0; i
< count
; i
++) {
280 static void __exit
dm_exit(void)
282 int i
= ARRAY_SIZE(_exits
);
288 * Should be empty by this point.
290 idr_destroy(&_minor_idr
);
294 * Block device functions
296 int dm_deleting_md(struct mapped_device
*md
)
298 return test_bit(DMF_DELETING
, &md
->flags
);
301 static int dm_blk_open(struct block_device
*bdev
, fmode_t mode
)
303 struct mapped_device
*md
;
305 spin_lock(&_minor_lock
);
307 md
= bdev
->bd_disk
->private_data
;
311 if (test_bit(DMF_FREEING
, &md
->flags
) ||
312 dm_deleting_md(md
)) {
318 atomic_inc(&md
->open_count
);
320 spin_unlock(&_minor_lock
);
322 return md
? 0 : -ENXIO
;
325 static void dm_blk_close(struct gendisk
*disk
, fmode_t mode
)
327 struct mapped_device
*md
;
329 spin_lock(&_minor_lock
);
331 md
= disk
->private_data
;
335 if (atomic_dec_and_test(&md
->open_count
) &&
336 (test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
)))
337 queue_work(deferred_remove_workqueue
, &deferred_remove_work
);
341 spin_unlock(&_minor_lock
);
344 int dm_open_count(struct mapped_device
*md
)
346 return atomic_read(&md
->open_count
);
350 * Guarantees nothing is using the device before it's deleted.
352 int dm_lock_for_deletion(struct mapped_device
*md
, bool mark_deferred
, bool only_deferred
)
356 spin_lock(&_minor_lock
);
358 if (dm_open_count(md
)) {
361 set_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
362 } else if (only_deferred
&& !test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
))
365 set_bit(DMF_DELETING
, &md
->flags
);
367 spin_unlock(&_minor_lock
);
372 int dm_cancel_deferred_remove(struct mapped_device
*md
)
376 spin_lock(&_minor_lock
);
378 if (test_bit(DMF_DELETING
, &md
->flags
))
381 clear_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
383 spin_unlock(&_minor_lock
);
388 static void do_deferred_remove(struct work_struct
*w
)
390 dm_deferred_remove();
393 sector_t
dm_get_size(struct mapped_device
*md
)
395 return get_capacity(md
->disk
);
398 struct request_queue
*dm_get_md_queue(struct mapped_device
*md
)
403 struct dm_stats
*dm_get_stats(struct mapped_device
*md
)
408 static int dm_blk_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
410 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
412 return dm_get_geometry(md
, geo
);
415 static int dm_grab_bdev_for_ioctl(struct mapped_device
*md
,
416 struct block_device
**bdev
,
419 struct dm_target
*tgt
;
420 struct dm_table
*map
;
425 map
= dm_get_live_table(md
, &srcu_idx
);
426 if (!map
|| !dm_table_get_size(map
))
429 /* We only support devices that have a single target */
430 if (dm_table_get_num_targets(map
) != 1)
433 tgt
= dm_table_get_target(map
, 0);
434 if (!tgt
->type
->prepare_ioctl
)
437 if (dm_suspended_md(md
)) {
442 r
= tgt
->type
->prepare_ioctl(tgt
, bdev
, mode
);
447 dm_put_live_table(md
, srcu_idx
);
451 dm_put_live_table(md
, srcu_idx
);
452 if (r
== -ENOTCONN
&& !fatal_signal_pending(current
)) {
459 static int dm_blk_ioctl(struct block_device
*bdev
, fmode_t mode
,
460 unsigned int cmd
, unsigned long arg
)
462 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
465 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
471 * Target determined this ioctl is being issued against a
472 * subset of the parent bdev; require extra privileges.
474 if (!capable(CAP_SYS_RAWIO
)) {
476 "%s: sending ioctl %x to DM device without required privilege.",
483 r
= __blkdev_driver_ioctl(bdev
, mode
, cmd
, arg
);
489 static struct dm_io
*alloc_io(struct mapped_device
*md
)
491 return mempool_alloc(md
->io_pool
, GFP_NOIO
);
494 static void free_io(struct mapped_device
*md
, struct dm_io
*io
)
496 mempool_free(io
, md
->io_pool
);
499 static void free_tio(struct dm_target_io
*tio
)
501 bio_put(&tio
->clone
);
504 int md_in_flight(struct mapped_device
*md
)
506 return atomic_read(&md
->pending
[READ
]) +
507 atomic_read(&md
->pending
[WRITE
]);
510 static void start_io_acct(struct dm_io
*io
)
512 struct mapped_device
*md
= io
->md
;
513 struct bio
*bio
= io
->bio
;
515 int rw
= bio_data_dir(bio
);
517 io
->start_time
= jiffies
;
519 cpu
= part_stat_lock();
520 part_round_stats(md
->queue
, cpu
, &dm_disk(md
)->part0
);
522 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
],
523 atomic_inc_return(&md
->pending
[rw
]));
525 if (unlikely(dm_stats_used(&md
->stats
)))
526 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
527 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
528 false, 0, &io
->stats_aux
);
531 static void end_io_acct(struct dm_io
*io
)
533 struct mapped_device
*md
= io
->md
;
534 struct bio
*bio
= io
->bio
;
535 unsigned long duration
= jiffies
- io
->start_time
;
537 int rw
= bio_data_dir(bio
);
539 generic_end_io_acct(md
->queue
, rw
, &dm_disk(md
)->part0
, io
->start_time
);
541 if (unlikely(dm_stats_used(&md
->stats
)))
542 dm_stats_account_io(&md
->stats
, bio_data_dir(bio
),
543 bio
->bi_iter
.bi_sector
, bio_sectors(bio
),
544 true, duration
, &io
->stats_aux
);
547 * After this is decremented the bio must not be touched if it is
550 pending
= atomic_dec_return(&md
->pending
[rw
]);
551 atomic_set(&dm_disk(md
)->part0
.in_flight
[rw
], pending
);
552 pending
+= atomic_read(&md
->pending
[rw
^0x1]);
554 /* nudge anyone waiting on suspend queue */
560 * Add the bio to the list of deferred io.
562 static void queue_io(struct mapped_device
*md
, struct bio
*bio
)
566 spin_lock_irqsave(&md
->deferred_lock
, flags
);
567 bio_list_add(&md
->deferred
, bio
);
568 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
569 queue_work(md
->wq
, &md
->work
);
573 * Everyone (including functions in this file), should use this
574 * function to access the md->map field, and make sure they call
575 * dm_put_live_table() when finished.
577 struct dm_table
*dm_get_live_table(struct mapped_device
*md
, int *srcu_idx
) __acquires(md
->io_barrier
)
579 *srcu_idx
= srcu_read_lock(&md
->io_barrier
);
581 return srcu_dereference(md
->map
, &md
->io_barrier
);
584 void dm_put_live_table(struct mapped_device
*md
, int srcu_idx
) __releases(md
->io_barrier
)
586 srcu_read_unlock(&md
->io_barrier
, srcu_idx
);
589 void dm_sync_table(struct mapped_device
*md
)
591 synchronize_srcu(&md
->io_barrier
);
592 synchronize_rcu_expedited();
596 * A fast alternative to dm_get_live_table/dm_put_live_table.
597 * The caller must not block between these two functions.
599 static struct dm_table
*dm_get_live_table_fast(struct mapped_device
*md
) __acquires(RCU
)
602 return rcu_dereference(md
->map
);
605 static void dm_put_live_table_fast(struct mapped_device
*md
) __releases(RCU
)
611 * Open a table device so we can use it as a map destination.
613 static int open_table_device(struct table_device
*td
, dev_t dev
,
614 struct mapped_device
*md
)
616 static char *_claim_ptr
= "I belong to device-mapper";
617 struct block_device
*bdev
;
621 BUG_ON(td
->dm_dev
.bdev
);
623 bdev
= blkdev_get_by_dev(dev
, td
->dm_dev
.mode
| FMODE_EXCL
, _claim_ptr
);
625 return PTR_ERR(bdev
);
627 r
= bd_link_disk_holder(bdev
, dm_disk(md
));
629 blkdev_put(bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
633 td
->dm_dev
.bdev
= bdev
;
634 td
->dm_dev
.dax_dev
= dax_get_by_host(bdev
->bd_disk
->disk_name
);
639 * Close a table device that we've been using.
641 static void close_table_device(struct table_device
*td
, struct mapped_device
*md
)
643 if (!td
->dm_dev
.bdev
)
646 bd_unlink_disk_holder(td
->dm_dev
.bdev
, dm_disk(md
));
647 blkdev_put(td
->dm_dev
.bdev
, td
->dm_dev
.mode
| FMODE_EXCL
);
648 put_dax(td
->dm_dev
.dax_dev
);
649 td
->dm_dev
.bdev
= NULL
;
650 td
->dm_dev
.dax_dev
= NULL
;
653 static struct table_device
*find_table_device(struct list_head
*l
, dev_t dev
,
655 struct table_device
*td
;
657 list_for_each_entry(td
, l
, list
)
658 if (td
->dm_dev
.bdev
->bd_dev
== dev
&& td
->dm_dev
.mode
== mode
)
664 int dm_get_table_device(struct mapped_device
*md
, dev_t dev
, fmode_t mode
,
665 struct dm_dev
**result
) {
667 struct table_device
*td
;
669 mutex_lock(&md
->table_devices_lock
);
670 td
= find_table_device(&md
->table_devices
, dev
, mode
);
672 td
= kmalloc_node(sizeof(*td
), GFP_KERNEL
, md
->numa_node_id
);
674 mutex_unlock(&md
->table_devices_lock
);
678 td
->dm_dev
.mode
= mode
;
679 td
->dm_dev
.bdev
= NULL
;
681 if ((r
= open_table_device(td
, dev
, md
))) {
682 mutex_unlock(&md
->table_devices_lock
);
687 format_dev_t(td
->dm_dev
.name
, dev
);
689 refcount_set(&td
->count
, 1);
690 list_add(&td
->list
, &md
->table_devices
);
692 refcount_inc(&td
->count
);
694 mutex_unlock(&md
->table_devices_lock
);
696 *result
= &td
->dm_dev
;
699 EXPORT_SYMBOL_GPL(dm_get_table_device
);
701 void dm_put_table_device(struct mapped_device
*md
, struct dm_dev
*d
)
703 struct table_device
*td
= container_of(d
, struct table_device
, dm_dev
);
705 mutex_lock(&md
->table_devices_lock
);
706 if (refcount_dec_and_test(&td
->count
)) {
707 close_table_device(td
, md
);
711 mutex_unlock(&md
->table_devices_lock
);
713 EXPORT_SYMBOL(dm_put_table_device
);
715 static void free_table_devices(struct list_head
*devices
)
717 struct list_head
*tmp
, *next
;
719 list_for_each_safe(tmp
, next
, devices
) {
720 struct table_device
*td
= list_entry(tmp
, struct table_device
, list
);
722 DMWARN("dm_destroy: %s still exists with %d references",
723 td
->dm_dev
.name
, refcount_read(&td
->count
));
729 * Get the geometry associated with a dm device
731 int dm_get_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
739 * Set the geometry of a device.
741 int dm_set_geometry(struct mapped_device
*md
, struct hd_geometry
*geo
)
743 sector_t sz
= (sector_t
)geo
->cylinders
* geo
->heads
* geo
->sectors
;
745 if (geo
->start
> sz
) {
746 DMWARN("Start sector is beyond the geometry limits.");
755 /*-----------------------------------------------------------------
757 * A more elegant soln is in the works that uses the queue
758 * merge fn, unfortunately there are a couple of changes to
759 * the block layer that I want to make for this. So in the
760 * interests of getting something for people to use I give
761 * you this clearly demarcated crap.
762 *---------------------------------------------------------------*/
764 static int __noflush_suspending(struct mapped_device
*md
)
766 return test_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
770 * Decrements the number of outstanding ios that a bio has been
771 * cloned into, completing the original io if necc.
773 static void dec_pending(struct dm_io
*io
, blk_status_t error
)
776 blk_status_t io_error
;
778 struct mapped_device
*md
= io
->md
;
780 /* Push-back supersedes any I/O errors */
781 if (unlikely(error
)) {
782 spin_lock_irqsave(&io
->endio_lock
, flags
);
783 if (!(io
->status
== BLK_STS_DM_REQUEUE
&&
784 __noflush_suspending(md
)))
786 spin_unlock_irqrestore(&io
->endio_lock
, flags
);
789 if (atomic_dec_and_test(&io
->io_count
)) {
790 if (io
->status
== BLK_STS_DM_REQUEUE
) {
792 * Target requested pushing back the I/O.
794 spin_lock_irqsave(&md
->deferred_lock
, flags
);
795 if (__noflush_suspending(md
))
796 bio_list_add_head(&md
->deferred
, io
->bio
);
798 /* noflush suspend was interrupted. */
799 io
->status
= BLK_STS_IOERR
;
800 spin_unlock_irqrestore(&md
->deferred_lock
, flags
);
803 io_error
= io
->status
;
808 if (io_error
== BLK_STS_DM_REQUEUE
)
811 if ((bio
->bi_opf
& REQ_PREFLUSH
) && bio
->bi_iter
.bi_size
) {
813 * Preflush done for flush with data, reissue
814 * without REQ_PREFLUSH.
816 bio
->bi_opf
&= ~REQ_PREFLUSH
;
819 /* done with normal IO or empty flush */
820 bio
->bi_status
= io_error
;
826 void disable_write_same(struct mapped_device
*md
)
828 struct queue_limits
*limits
= dm_get_queue_limits(md
);
830 /* device doesn't really support WRITE SAME, disable it */
831 limits
->max_write_same_sectors
= 0;
834 void disable_write_zeroes(struct mapped_device
*md
)
836 struct queue_limits
*limits
= dm_get_queue_limits(md
);
838 /* device doesn't really support WRITE ZEROES, disable it */
839 limits
->max_write_zeroes_sectors
= 0;
842 static void clone_endio(struct bio
*bio
)
844 blk_status_t error
= bio
->bi_status
;
845 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
846 struct dm_io
*io
= tio
->io
;
847 struct mapped_device
*md
= tio
->io
->md
;
848 dm_endio_fn endio
= tio
->ti
->type
->end_io
;
850 if (unlikely(error
== BLK_STS_TARGET
)) {
851 if (bio_op(bio
) == REQ_OP_WRITE_SAME
&&
852 !bio
->bi_disk
->queue
->limits
.max_write_same_sectors
)
853 disable_write_same(md
);
854 if (bio_op(bio
) == REQ_OP_WRITE_ZEROES
&&
855 !bio
->bi_disk
->queue
->limits
.max_write_zeroes_sectors
)
856 disable_write_zeroes(md
);
860 int r
= endio(tio
->ti
, bio
, &error
);
862 case DM_ENDIO_REQUEUE
:
863 error
= BLK_STS_DM_REQUEUE
;
867 case DM_ENDIO_INCOMPLETE
:
868 /* The target will handle the io */
871 DMWARN("unimplemented target endio return value: %d", r
);
877 dec_pending(io
, error
);
881 * Return maximum size of I/O possible at the supplied sector up to the current
884 static sector_t
max_io_len_target_boundary(sector_t sector
, struct dm_target
*ti
)
886 sector_t target_offset
= dm_target_offset(ti
, sector
);
888 return ti
->len
- target_offset
;
891 static sector_t
max_io_len(sector_t sector
, struct dm_target
*ti
)
893 sector_t len
= max_io_len_target_boundary(sector
, ti
);
894 sector_t offset
, max_len
;
897 * Does the target need to split even further?
899 if (ti
->max_io_len
) {
900 offset
= dm_target_offset(ti
, sector
);
901 if (unlikely(ti
->max_io_len
& (ti
->max_io_len
- 1)))
902 max_len
= sector_div(offset
, ti
->max_io_len
);
904 max_len
= offset
& (ti
->max_io_len
- 1);
905 max_len
= ti
->max_io_len
- max_len
;
914 int dm_set_target_max_io_len(struct dm_target
*ti
, sector_t len
)
916 if (len
> UINT_MAX
) {
917 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
918 (unsigned long long)len
, UINT_MAX
);
919 ti
->error
= "Maximum size of target IO is too large";
923 ti
->max_io_len
= (uint32_t) len
;
927 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len
);
929 static struct dm_target
*dm_dax_get_live_target(struct mapped_device
*md
,
930 sector_t sector
, int *srcu_idx
)
932 struct dm_table
*map
;
933 struct dm_target
*ti
;
935 map
= dm_get_live_table(md
, srcu_idx
);
939 ti
= dm_table_find_target(map
, sector
);
940 if (!dm_target_is_valid(ti
))
946 static long dm_dax_direct_access(struct dax_device
*dax_dev
, pgoff_t pgoff
,
947 long nr_pages
, void **kaddr
, pfn_t
*pfn
)
949 struct mapped_device
*md
= dax_get_private(dax_dev
);
950 sector_t sector
= pgoff
* PAGE_SECTORS
;
951 struct dm_target
*ti
;
952 long len
, ret
= -EIO
;
955 ti
= dm_dax_get_live_target(md
, sector
, &srcu_idx
);
959 if (!ti
->type
->direct_access
)
961 len
= max_io_len(sector
, ti
) / PAGE_SECTORS
;
964 nr_pages
= min(len
, nr_pages
);
965 if (ti
->type
->direct_access
)
966 ret
= ti
->type
->direct_access(ti
, pgoff
, nr_pages
, kaddr
, pfn
);
969 dm_put_live_table(md
, srcu_idx
);
974 static size_t dm_dax_copy_from_iter(struct dax_device
*dax_dev
, pgoff_t pgoff
,
975 void *addr
, size_t bytes
, struct iov_iter
*i
)
977 struct mapped_device
*md
= dax_get_private(dax_dev
);
978 sector_t sector
= pgoff
* PAGE_SECTORS
;
979 struct dm_target
*ti
;
983 ti
= dm_dax_get_live_target(md
, sector
, &srcu_idx
);
987 if (!ti
->type
->dax_copy_from_iter
) {
988 ret
= copy_from_iter(addr
, bytes
, i
);
991 ret
= ti
->type
->dax_copy_from_iter(ti
, pgoff
, addr
, bytes
, i
);
993 dm_put_live_table(md
, srcu_idx
);
999 * A target may call dm_accept_partial_bio only from the map routine. It is
1000 * allowed for all bio types except REQ_PREFLUSH.
1002 * dm_accept_partial_bio informs the dm that the target only wants to process
1003 * additional n_sectors sectors of the bio and the rest of the data should be
1004 * sent in a next bio.
1006 * A diagram that explains the arithmetics:
1007 * +--------------------+---------------+-------+
1009 * +--------------------+---------------+-------+
1011 * <-------------- *tio->len_ptr --------------->
1012 * <------- bi_size ------->
1015 * Region 1 was already iterated over with bio_advance or similar function.
1016 * (it may be empty if the target doesn't use bio_advance)
1017 * Region 2 is the remaining bio size that the target wants to process.
1018 * (it may be empty if region 1 is non-empty, although there is no reason
1020 * The target requires that region 3 is to be sent in the next bio.
1022 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1023 * the partially processed part (the sum of regions 1+2) must be the same for all
1024 * copies of the bio.
1026 void dm_accept_partial_bio(struct bio
*bio
, unsigned n_sectors
)
1028 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1029 unsigned bi_size
= bio
->bi_iter
.bi_size
>> SECTOR_SHIFT
;
1030 BUG_ON(bio
->bi_opf
& REQ_PREFLUSH
);
1031 BUG_ON(bi_size
> *tio
->len_ptr
);
1032 BUG_ON(n_sectors
> bi_size
);
1033 *tio
->len_ptr
-= bi_size
- n_sectors
;
1034 bio
->bi_iter
.bi_size
= n_sectors
<< SECTOR_SHIFT
;
1036 EXPORT_SYMBOL_GPL(dm_accept_partial_bio
);
1039 * The zone descriptors obtained with a zone report indicate
1040 * zone positions within the target device. The zone descriptors
1041 * must be remapped to match their position within the dm device.
1042 * A target may call dm_remap_zone_report after completion of a
1043 * REQ_OP_ZONE_REPORT bio to remap the zone descriptors obtained
1044 * from the target device mapping to the dm device.
1046 void dm_remap_zone_report(struct dm_target
*ti
, struct bio
*bio
, sector_t start
)
1048 #ifdef CONFIG_BLK_DEV_ZONED
1049 struct dm_target_io
*tio
= container_of(bio
, struct dm_target_io
, clone
);
1050 struct bio
*report_bio
= tio
->io
->bio
;
1051 struct blk_zone_report_hdr
*hdr
= NULL
;
1052 struct blk_zone
*zone
;
1053 unsigned int nr_rep
= 0;
1055 struct bio_vec bvec
;
1056 struct bvec_iter iter
;
1063 * Remap the start sector of the reported zones. For sequential zones,
1064 * also remap the write pointer position.
1066 bio_for_each_segment(bvec
, report_bio
, iter
) {
1067 addr
= kmap_atomic(bvec
.bv_page
);
1069 /* Remember the report header in the first page */
1072 ofst
= sizeof(struct blk_zone_report_hdr
);
1076 /* Set zones start sector */
1077 while (hdr
->nr_zones
&& ofst
< bvec
.bv_len
) {
1079 if (zone
->start
>= start
+ ti
->len
) {
1083 zone
->start
= zone
->start
+ ti
->begin
- start
;
1084 if (zone
->type
!= BLK_ZONE_TYPE_CONVENTIONAL
) {
1085 if (zone
->cond
== BLK_ZONE_COND_FULL
)
1086 zone
->wp
= zone
->start
+ zone
->len
;
1087 else if (zone
->cond
== BLK_ZONE_COND_EMPTY
)
1088 zone
->wp
= zone
->start
;
1090 zone
->wp
= zone
->wp
+ ti
->begin
- start
;
1092 ofst
+= sizeof(struct blk_zone
);
1098 kunmap_atomic(addr
);
1105 hdr
->nr_zones
= nr_rep
;
1109 bio_advance(report_bio
, report_bio
->bi_iter
.bi_size
);
1111 #else /* !CONFIG_BLK_DEV_ZONED */
1112 bio
->bi_status
= BLK_STS_NOTSUPP
;
1115 EXPORT_SYMBOL_GPL(dm_remap_zone_report
);
1118 * Flush current->bio_list when the target map method blocks.
1119 * This fixes deadlocks in snapshot and possibly in other targets.
1122 struct blk_plug plug
;
1123 struct blk_plug_cb cb
;
1126 static void flush_current_bio_list(struct blk_plug_cb
*cb
, bool from_schedule
)
1128 struct dm_offload
*o
= container_of(cb
, struct dm_offload
, cb
);
1129 struct bio_list list
;
1133 INIT_LIST_HEAD(&o
->cb
.list
);
1135 if (unlikely(!current
->bio_list
))
1138 for (i
= 0; i
< 2; i
++) {
1139 list
= current
->bio_list
[i
];
1140 bio_list_init(¤t
->bio_list
[i
]);
1142 while ((bio
= bio_list_pop(&list
))) {
1143 struct bio_set
*bs
= bio
->bi_pool
;
1144 if (unlikely(!bs
) || bs
== fs_bio_set
||
1145 !bs
->rescue_workqueue
) {
1146 bio_list_add(¤t
->bio_list
[i
], bio
);
1150 spin_lock(&bs
->rescue_lock
);
1151 bio_list_add(&bs
->rescue_list
, bio
);
1152 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
1153 spin_unlock(&bs
->rescue_lock
);
1158 static void dm_offload_start(struct dm_offload
*o
)
1160 blk_start_plug(&o
->plug
);
1161 o
->cb
.callback
= flush_current_bio_list
;
1162 list_add(&o
->cb
.list
, ¤t
->plug
->cb_list
);
1165 static void dm_offload_end(struct dm_offload
*o
)
1167 list_del(&o
->cb
.list
);
1168 blk_finish_plug(&o
->plug
);
1171 static void __map_bio(struct dm_target_io
*tio
)
1175 struct dm_offload o
;
1176 struct bio
*clone
= &tio
->clone
;
1177 struct dm_target
*ti
= tio
->ti
;
1179 clone
->bi_end_io
= clone_endio
;
1182 * Map the clone. If r == 0 we don't need to do
1183 * anything, the target has assumed ownership of
1186 atomic_inc(&tio
->io
->io_count
);
1187 sector
= clone
->bi_iter
.bi_sector
;
1189 dm_offload_start(&o
);
1190 r
= ti
->type
->map(ti
, clone
);
1194 case DM_MAPIO_SUBMITTED
:
1196 case DM_MAPIO_REMAPPED
:
1197 /* the bio has been remapped so dispatch it */
1198 trace_block_bio_remap(clone
->bi_disk
->queue
, clone
,
1199 bio_dev(tio
->io
->bio
), sector
);
1200 generic_make_request(clone
);
1203 dec_pending(tio
->io
, BLK_STS_IOERR
);
1206 case DM_MAPIO_REQUEUE
:
1207 dec_pending(tio
->io
, BLK_STS_DM_REQUEUE
);
1211 DMWARN("unimplemented target map return value: %d", r
);
1217 struct mapped_device
*md
;
1218 struct dm_table
*map
;
1222 unsigned sector_count
;
1225 static void bio_setup_sector(struct bio
*bio
, sector_t sector
, unsigned len
)
1227 bio
->bi_iter
.bi_sector
= sector
;
1228 bio
->bi_iter
.bi_size
= to_bytes(len
);
1232 * Creates a bio that consists of range of complete bvecs.
1234 static int clone_bio(struct dm_target_io
*tio
, struct bio
*bio
,
1235 sector_t sector
, unsigned len
)
1237 struct bio
*clone
= &tio
->clone
;
1239 __bio_clone_fast(clone
, bio
);
1241 if (unlikely(bio_integrity(bio
) != NULL
)) {
1244 if (unlikely(!dm_target_has_integrity(tio
->ti
->type
) &&
1245 !dm_target_passes_integrity(tio
->ti
->type
))) {
1246 DMWARN("%s: the target %s doesn't support integrity data.",
1247 dm_device_name(tio
->io
->md
),
1248 tio
->ti
->type
->name
);
1252 r
= bio_integrity_clone(clone
, bio
, GFP_NOIO
);
1257 if (bio_op(bio
) != REQ_OP_ZONE_REPORT
)
1258 bio_advance(clone
, to_bytes(sector
- clone
->bi_iter
.bi_sector
));
1259 clone
->bi_iter
.bi_size
= to_bytes(len
);
1261 if (unlikely(bio_integrity(bio
) != NULL
))
1262 bio_integrity_trim(clone
);
1267 static struct dm_target_io
*alloc_tio(struct clone_info
*ci
,
1268 struct dm_target
*ti
,
1269 unsigned target_bio_nr
)
1271 struct dm_target_io
*tio
;
1274 clone
= bio_alloc_bioset(GFP_NOIO
, 0, ci
->md
->bs
);
1275 tio
= container_of(clone
, struct dm_target_io
, clone
);
1279 tio
->target_bio_nr
= target_bio_nr
;
1284 static void __clone_and_map_simple_bio(struct clone_info
*ci
,
1285 struct dm_target
*ti
,
1286 unsigned target_bio_nr
, unsigned *len
)
1288 struct dm_target_io
*tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1289 struct bio
*clone
= &tio
->clone
;
1293 __bio_clone_fast(clone
, ci
->bio
);
1295 bio_setup_sector(clone
, ci
->sector
, *len
);
1300 static void __send_duplicate_bios(struct clone_info
*ci
, struct dm_target
*ti
,
1301 unsigned num_bios
, unsigned *len
)
1303 unsigned target_bio_nr
;
1305 for (target_bio_nr
= 0; target_bio_nr
< num_bios
; target_bio_nr
++)
1306 __clone_and_map_simple_bio(ci
, ti
, target_bio_nr
, len
);
1309 static int __send_empty_flush(struct clone_info
*ci
)
1311 unsigned target_nr
= 0;
1312 struct dm_target
*ti
;
1314 BUG_ON(bio_has_data(ci
->bio
));
1315 while ((ti
= dm_table_get_target(ci
->map
, target_nr
++)))
1316 __send_duplicate_bios(ci
, ti
, ti
->num_flush_bios
, NULL
);
1321 static int __clone_and_map_data_bio(struct clone_info
*ci
, struct dm_target
*ti
,
1322 sector_t sector
, unsigned *len
)
1324 struct bio
*bio
= ci
->bio
;
1325 struct dm_target_io
*tio
;
1326 unsigned target_bio_nr
;
1327 unsigned num_target_bios
= 1;
1331 * Does the target want to receive duplicate copies of the bio?
1333 if (bio_data_dir(bio
) == WRITE
&& ti
->num_write_bios
)
1334 num_target_bios
= ti
->num_write_bios(ti
, bio
);
1336 for (target_bio_nr
= 0; target_bio_nr
< num_target_bios
; target_bio_nr
++) {
1337 tio
= alloc_tio(ci
, ti
, target_bio_nr
);
1339 r
= clone_bio(tio
, bio
, sector
, *len
);
1350 typedef unsigned (*get_num_bios_fn
)(struct dm_target
*ti
);
1352 static unsigned get_num_discard_bios(struct dm_target
*ti
)
1354 return ti
->num_discard_bios
;
1357 static unsigned get_num_write_same_bios(struct dm_target
*ti
)
1359 return ti
->num_write_same_bios
;
1362 static unsigned get_num_write_zeroes_bios(struct dm_target
*ti
)
1364 return ti
->num_write_zeroes_bios
;
1367 typedef bool (*is_split_required_fn
)(struct dm_target
*ti
);
1369 static bool is_split_required_for_discard(struct dm_target
*ti
)
1371 return ti
->split_discard_bios
;
1374 static int __send_changing_extent_only(struct clone_info
*ci
,
1375 get_num_bios_fn get_num_bios
,
1376 is_split_required_fn is_split_required
)
1378 struct dm_target
*ti
;
1383 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1384 if (!dm_target_is_valid(ti
))
1388 * Even though the device advertised support for this type of
1389 * request, that does not mean every target supports it, and
1390 * reconfiguration might also have changed that since the
1391 * check was performed.
1393 num_bios
= get_num_bios
? get_num_bios(ti
) : 0;
1397 if (is_split_required
&& !is_split_required(ti
))
1398 len
= min((sector_t
)ci
->sector_count
, max_io_len_target_boundary(ci
->sector
, ti
));
1400 len
= min((sector_t
)ci
->sector_count
, max_io_len(ci
->sector
, ti
));
1402 __send_duplicate_bios(ci
, ti
, num_bios
, &len
);
1405 } while (ci
->sector_count
-= len
);
1410 static int __send_discard(struct clone_info
*ci
)
1412 return __send_changing_extent_only(ci
, get_num_discard_bios
,
1413 is_split_required_for_discard
);
1416 static int __send_write_same(struct clone_info
*ci
)
1418 return __send_changing_extent_only(ci
, get_num_write_same_bios
, NULL
);
1421 static int __send_write_zeroes(struct clone_info
*ci
)
1423 return __send_changing_extent_only(ci
, get_num_write_zeroes_bios
, NULL
);
1427 * Select the correct strategy for processing a non-flush bio.
1429 static int __split_and_process_non_flush(struct clone_info
*ci
)
1431 struct bio
*bio
= ci
->bio
;
1432 struct dm_target
*ti
;
1436 if (unlikely(bio_op(bio
) == REQ_OP_DISCARD
))
1437 return __send_discard(ci
);
1438 else if (unlikely(bio_op(bio
) == REQ_OP_WRITE_SAME
))
1439 return __send_write_same(ci
);
1440 else if (unlikely(bio_op(bio
) == REQ_OP_WRITE_ZEROES
))
1441 return __send_write_zeroes(ci
);
1443 ti
= dm_table_find_target(ci
->map
, ci
->sector
);
1444 if (!dm_target_is_valid(ti
))
1447 if (bio_op(bio
) == REQ_OP_ZONE_REPORT
)
1448 len
= ci
->sector_count
;
1450 len
= min_t(sector_t
, max_io_len(ci
->sector
, ti
),
1453 r
= __clone_and_map_data_bio(ci
, ti
, ci
->sector
, &len
);
1458 ci
->sector_count
-= len
;
1464 * Entry point to split a bio into clones and submit them to the targets.
1466 static void __split_and_process_bio(struct mapped_device
*md
,
1467 struct dm_table
*map
, struct bio
*bio
)
1469 struct clone_info ci
;
1472 if (unlikely(!map
)) {
1479 ci
.io
= alloc_io(md
);
1481 atomic_set(&ci
.io
->io_count
, 1);
1484 spin_lock_init(&ci
.io
->endio_lock
);
1485 ci
.sector
= bio
->bi_iter
.bi_sector
;
1487 start_io_acct(ci
.io
);
1489 if (bio
->bi_opf
& REQ_PREFLUSH
) {
1490 ci
.bio
= &ci
.md
->flush_bio
;
1491 ci
.sector_count
= 0;
1492 error
= __send_empty_flush(&ci
);
1493 /* dec_pending submits any data associated with flush */
1494 } else if (bio_op(bio
) == REQ_OP_ZONE_RESET
) {
1496 ci
.sector_count
= 0;
1497 error
= __split_and_process_non_flush(&ci
);
1500 ci
.sector_count
= bio_sectors(bio
);
1501 while (ci
.sector_count
&& !error
)
1502 error
= __split_and_process_non_flush(&ci
);
1505 /* drop the extra reference count */
1506 dec_pending(ci
.io
, errno_to_blk_status(error
));
1508 /*-----------------------------------------------------------------
1510 *---------------------------------------------------------------*/
1513 * The request function that just remaps the bio built up by
1516 static blk_qc_t
dm_make_request(struct request_queue
*q
, struct bio
*bio
)
1518 int rw
= bio_data_dir(bio
);
1519 struct mapped_device
*md
= q
->queuedata
;
1521 struct dm_table
*map
;
1523 map
= dm_get_live_table(md
, &srcu_idx
);
1525 generic_start_io_acct(q
, rw
, bio_sectors(bio
), &dm_disk(md
)->part0
);
1527 /* if we're suspended, we have to queue this io for later */
1528 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
))) {
1529 dm_put_live_table(md
, srcu_idx
);
1531 if (!(bio
->bi_opf
& REQ_RAHEAD
))
1535 return BLK_QC_T_NONE
;
1538 __split_and_process_bio(md
, map
, bio
);
1539 dm_put_live_table(md
, srcu_idx
);
1540 return BLK_QC_T_NONE
;
1543 static int dm_any_congested(void *congested_data
, int bdi_bits
)
1546 struct mapped_device
*md
= congested_data
;
1547 struct dm_table
*map
;
1549 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
1550 if (dm_request_based(md
)) {
1552 * With request-based DM we only need to check the
1553 * top-level queue for congestion.
1555 r
= md
->queue
->backing_dev_info
->wb
.state
& bdi_bits
;
1557 map
= dm_get_live_table_fast(md
);
1559 r
= dm_table_any_congested(map
, bdi_bits
);
1560 dm_put_live_table_fast(md
);
1567 /*-----------------------------------------------------------------
1568 * An IDR is used to keep track of allocated minor numbers.
1569 *---------------------------------------------------------------*/
1570 static void free_minor(int minor
)
1572 spin_lock(&_minor_lock
);
1573 idr_remove(&_minor_idr
, minor
);
1574 spin_unlock(&_minor_lock
);
1578 * See if the device with a specific minor # is free.
1580 static int specific_minor(int minor
)
1584 if (minor
>= (1 << MINORBITS
))
1587 idr_preload(GFP_KERNEL
);
1588 spin_lock(&_minor_lock
);
1590 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, minor
, minor
+ 1, GFP_NOWAIT
);
1592 spin_unlock(&_minor_lock
);
1595 return r
== -ENOSPC
? -EBUSY
: r
;
1599 static int next_free_minor(int *minor
)
1603 idr_preload(GFP_KERNEL
);
1604 spin_lock(&_minor_lock
);
1606 r
= idr_alloc(&_minor_idr
, MINOR_ALLOCED
, 0, 1 << MINORBITS
, GFP_NOWAIT
);
1608 spin_unlock(&_minor_lock
);
1616 static const struct block_device_operations dm_blk_dops
;
1617 static const struct dax_operations dm_dax_ops
;
1619 static void dm_wq_work(struct work_struct
*work
);
1621 void dm_init_md_queue(struct mapped_device
*md
)
1624 * Initialize data that will only be used by a non-blk-mq DM queue
1625 * - must do so here (in alloc_dev callchain) before queue is used
1627 md
->queue
->queuedata
= md
;
1628 md
->queue
->backing_dev_info
->congested_data
= md
;
1631 void dm_init_normal_md_queue(struct mapped_device
*md
)
1633 md
->use_blk_mq
= false;
1634 dm_init_md_queue(md
);
1637 * Initialize aspects of queue that aren't relevant for blk-mq
1639 md
->queue
->backing_dev_info
->congested_fn
= dm_any_congested
;
1642 static void cleanup_mapped_device(struct mapped_device
*md
)
1645 destroy_workqueue(md
->wq
);
1646 if (md
->kworker_task
)
1647 kthread_stop(md
->kworker_task
);
1648 mempool_destroy(md
->io_pool
);
1650 bioset_free(md
->bs
);
1653 kill_dax(md
->dax_dev
);
1654 put_dax(md
->dax_dev
);
1659 spin_lock(&_minor_lock
);
1660 md
->disk
->private_data
= NULL
;
1661 spin_unlock(&_minor_lock
);
1662 del_gendisk(md
->disk
);
1667 blk_cleanup_queue(md
->queue
);
1669 cleanup_srcu_struct(&md
->io_barrier
);
1676 dm_mq_cleanup_mapped_device(md
);
1680 * Allocate and initialise a blank device with a given minor.
1682 static struct mapped_device
*alloc_dev(int minor
)
1684 int r
, numa_node_id
= dm_get_numa_node();
1685 struct dax_device
*dax_dev
;
1686 struct mapped_device
*md
;
1689 md
= kvzalloc_node(sizeof(*md
), GFP_KERNEL
, numa_node_id
);
1691 DMWARN("unable to allocate device, out of memory.");
1695 if (!try_module_get(THIS_MODULE
))
1696 goto bad_module_get
;
1698 /* get a minor number for the dev */
1699 if (minor
== DM_ANY_MINOR
)
1700 r
= next_free_minor(&minor
);
1702 r
= specific_minor(minor
);
1706 r
= init_srcu_struct(&md
->io_barrier
);
1708 goto bad_io_barrier
;
1710 md
->numa_node_id
= numa_node_id
;
1711 md
->use_blk_mq
= dm_use_blk_mq_default();
1712 md
->init_tio_pdu
= false;
1713 md
->type
= DM_TYPE_NONE
;
1714 mutex_init(&md
->suspend_lock
);
1715 mutex_init(&md
->type_lock
);
1716 mutex_init(&md
->table_devices_lock
);
1717 spin_lock_init(&md
->deferred_lock
);
1718 atomic_set(&md
->holders
, 1);
1719 atomic_set(&md
->open_count
, 0);
1720 atomic_set(&md
->event_nr
, 0);
1721 atomic_set(&md
->uevent_seq
, 0);
1722 INIT_LIST_HEAD(&md
->uevent_list
);
1723 INIT_LIST_HEAD(&md
->table_devices
);
1724 spin_lock_init(&md
->uevent_lock
);
1726 md
->queue
= blk_alloc_queue_node(GFP_KERNEL
, numa_node_id
);
1730 dm_init_md_queue(md
);
1732 md
->disk
= alloc_disk_node(1, numa_node_id
);
1736 atomic_set(&md
->pending
[0], 0);
1737 atomic_set(&md
->pending
[1], 0);
1738 init_waitqueue_head(&md
->wait
);
1739 INIT_WORK(&md
->work
, dm_wq_work
);
1740 init_waitqueue_head(&md
->eventq
);
1741 init_completion(&md
->kobj_holder
.completion
);
1742 md
->kworker_task
= NULL
;
1744 md
->disk
->major
= _major
;
1745 md
->disk
->first_minor
= minor
;
1746 md
->disk
->fops
= &dm_blk_dops
;
1747 md
->disk
->queue
= md
->queue
;
1748 md
->disk
->private_data
= md
;
1749 sprintf(md
->disk
->disk_name
, "dm-%d", minor
);
1751 dax_dev
= alloc_dax(md
, md
->disk
->disk_name
, &dm_dax_ops
);
1754 md
->dax_dev
= dax_dev
;
1757 format_dev_t(md
->name
, MKDEV(_major
, minor
));
1759 md
->wq
= alloc_workqueue("kdmflush", WQ_MEM_RECLAIM
, 0);
1763 md
->bdev
= bdget_disk(md
->disk
, 0);
1767 bio_init(&md
->flush_bio
, NULL
, 0);
1768 bio_set_dev(&md
->flush_bio
, md
->bdev
);
1769 md
->flush_bio
.bi_opf
= REQ_OP_WRITE
| REQ_PREFLUSH
| REQ_SYNC
;
1771 dm_stats_init(&md
->stats
);
1773 /* Populate the mapping, nobody knows we exist yet */
1774 spin_lock(&_minor_lock
);
1775 old_md
= idr_replace(&_minor_idr
, md
, minor
);
1776 spin_unlock(&_minor_lock
);
1778 BUG_ON(old_md
!= MINOR_ALLOCED
);
1783 cleanup_mapped_device(md
);
1787 module_put(THIS_MODULE
);
1793 static void unlock_fs(struct mapped_device
*md
);
1795 static void free_dev(struct mapped_device
*md
)
1797 int minor
= MINOR(disk_devt(md
->disk
));
1801 cleanup_mapped_device(md
);
1803 free_table_devices(&md
->table_devices
);
1804 dm_stats_cleanup(&md
->stats
);
1807 module_put(THIS_MODULE
);
1811 static void __bind_mempools(struct mapped_device
*md
, struct dm_table
*t
)
1813 struct dm_md_mempools
*p
= dm_table_get_md_mempools(t
);
1816 /* The md already has necessary mempools. */
1817 if (dm_table_bio_based(t
)) {
1819 * Reload bioset because front_pad may have changed
1820 * because a different table was loaded.
1822 bioset_free(md
->bs
);
1827 * There's no need to reload with request-based dm
1828 * because the size of front_pad doesn't change.
1829 * Note for future: If you are to reload bioset,
1830 * prep-ed requests in the queue may refer
1831 * to bio from the old bioset, so you must walk
1832 * through the queue to unprep.
1837 BUG_ON(!p
|| md
->io_pool
|| md
->bs
);
1839 md
->io_pool
= p
->io_pool
;
1845 /* mempool bind completed, no longer need any mempools in the table */
1846 dm_table_free_md_mempools(t
);
1850 * Bind a table to the device.
1852 static void event_callback(void *context
)
1854 unsigned long flags
;
1856 struct mapped_device
*md
= (struct mapped_device
*) context
;
1858 spin_lock_irqsave(&md
->uevent_lock
, flags
);
1859 list_splice_init(&md
->uevent_list
, &uevents
);
1860 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
1862 dm_send_uevents(&uevents
, &disk_to_dev(md
->disk
)->kobj
);
1864 atomic_inc(&md
->event_nr
);
1865 wake_up(&md
->eventq
);
1866 dm_issue_global_event();
1870 * Protected by md->suspend_lock obtained by dm_swap_table().
1872 static void __set_size(struct mapped_device
*md
, sector_t size
)
1874 lockdep_assert_held(&md
->suspend_lock
);
1876 set_capacity(md
->disk
, size
);
1878 i_size_write(md
->bdev
->bd_inode
, (loff_t
)size
<< SECTOR_SHIFT
);
1882 * Returns old map, which caller must destroy.
1884 static struct dm_table
*__bind(struct mapped_device
*md
, struct dm_table
*t
,
1885 struct queue_limits
*limits
)
1887 struct dm_table
*old_map
;
1888 struct request_queue
*q
= md
->queue
;
1891 lockdep_assert_held(&md
->suspend_lock
);
1893 size
= dm_table_get_size(t
);
1896 * Wipe any geometry if the size of the table changed.
1898 if (size
!= dm_get_size(md
))
1899 memset(&md
->geometry
, 0, sizeof(md
->geometry
));
1901 __set_size(md
, size
);
1903 dm_table_event_callback(t
, event_callback
, md
);
1906 * The queue hasn't been stopped yet, if the old table type wasn't
1907 * for request-based during suspension. So stop it to prevent
1908 * I/O mapping before resume.
1909 * This must be done before setting the queue restrictions,
1910 * because request-based dm may be run just after the setting.
1912 if (dm_table_request_based(t
)) {
1915 * Leverage the fact that request-based DM targets are
1916 * immutable singletons and establish md->immutable_target
1917 * - used to optimize both dm_request_fn and dm_mq_queue_rq
1919 md
->immutable_target
= dm_table_get_immutable_target(t
);
1922 __bind_mempools(md
, t
);
1924 old_map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
1925 rcu_assign_pointer(md
->map
, (void *)t
);
1926 md
->immutable_target_type
= dm_table_get_immutable_target_type(t
);
1928 dm_table_set_restrictions(t
, q
, limits
);
1936 * Returns unbound table for the caller to free.
1938 static struct dm_table
*__unbind(struct mapped_device
*md
)
1940 struct dm_table
*map
= rcu_dereference_protected(md
->map
, 1);
1945 dm_table_event_callback(map
, NULL
, NULL
);
1946 RCU_INIT_POINTER(md
->map
, NULL
);
1953 * Constructor for a new device.
1955 int dm_create(int minor
, struct mapped_device
**result
)
1957 struct mapped_device
*md
;
1959 md
= alloc_dev(minor
);
1970 * Functions to manage md->type.
1971 * All are required to hold md->type_lock.
1973 void dm_lock_md_type(struct mapped_device
*md
)
1975 mutex_lock(&md
->type_lock
);
1978 void dm_unlock_md_type(struct mapped_device
*md
)
1980 mutex_unlock(&md
->type_lock
);
1983 void dm_set_md_type(struct mapped_device
*md
, enum dm_queue_mode type
)
1985 BUG_ON(!mutex_is_locked(&md
->type_lock
));
1989 enum dm_queue_mode
dm_get_md_type(struct mapped_device
*md
)
1994 struct target_type
*dm_get_immutable_target_type(struct mapped_device
*md
)
1996 return md
->immutable_target_type
;
2000 * The queue_limits are only valid as long as you have a reference
2003 struct queue_limits
*dm_get_queue_limits(struct mapped_device
*md
)
2005 BUG_ON(!atomic_read(&md
->holders
));
2006 return &md
->queue
->limits
;
2008 EXPORT_SYMBOL_GPL(dm_get_queue_limits
);
2011 * Setup the DM device's queue based on md's type
2013 int dm_setup_md_queue(struct mapped_device
*md
, struct dm_table
*t
)
2016 enum dm_queue_mode type
= dm_get_md_type(md
);
2019 case DM_TYPE_REQUEST_BASED
:
2020 r
= dm_old_init_request_queue(md
, t
);
2022 DMERR("Cannot initialize queue for request-based mapped device");
2026 case DM_TYPE_MQ_REQUEST_BASED
:
2027 r
= dm_mq_init_request_queue(md
, t
);
2029 DMERR("Cannot initialize queue for request-based dm-mq mapped device");
2033 case DM_TYPE_BIO_BASED
:
2034 case DM_TYPE_DAX_BIO_BASED
:
2035 dm_init_normal_md_queue(md
);
2036 blk_queue_make_request(md
->queue
, dm_make_request
);
2038 * DM handles splitting bios as needed. Free the bio_split bioset
2039 * since it won't be used (saves 1 process per bio-based DM device).
2041 bioset_free(md
->queue
->bio_split
);
2042 md
->queue
->bio_split
= NULL
;
2044 if (type
== DM_TYPE_DAX_BIO_BASED
)
2045 queue_flag_set_unlocked(QUEUE_FLAG_DAX
, md
->queue
);
2055 struct mapped_device
*dm_get_md(dev_t dev
)
2057 struct mapped_device
*md
;
2058 unsigned minor
= MINOR(dev
);
2060 if (MAJOR(dev
) != _major
|| minor
>= (1 << MINORBITS
))
2063 spin_lock(&_minor_lock
);
2065 md
= idr_find(&_minor_idr
, minor
);
2066 if (!md
|| md
== MINOR_ALLOCED
|| (MINOR(disk_devt(dm_disk(md
))) != minor
) ||
2067 test_bit(DMF_FREEING
, &md
->flags
) || dm_deleting_md(md
)) {
2073 spin_unlock(&_minor_lock
);
2077 EXPORT_SYMBOL_GPL(dm_get_md
);
2079 void *dm_get_mdptr(struct mapped_device
*md
)
2081 return md
->interface_ptr
;
2084 void dm_set_mdptr(struct mapped_device
*md
, void *ptr
)
2086 md
->interface_ptr
= ptr
;
2089 void dm_get(struct mapped_device
*md
)
2091 atomic_inc(&md
->holders
);
2092 BUG_ON(test_bit(DMF_FREEING
, &md
->flags
));
2095 int dm_hold(struct mapped_device
*md
)
2097 spin_lock(&_minor_lock
);
2098 if (test_bit(DMF_FREEING
, &md
->flags
)) {
2099 spin_unlock(&_minor_lock
);
2103 spin_unlock(&_minor_lock
);
2106 EXPORT_SYMBOL_GPL(dm_hold
);
2108 const char *dm_device_name(struct mapped_device
*md
)
2112 EXPORT_SYMBOL_GPL(dm_device_name
);
2114 static void __dm_destroy(struct mapped_device
*md
, bool wait
)
2116 struct request_queue
*q
= dm_get_md_queue(md
);
2117 struct dm_table
*map
;
2122 spin_lock(&_minor_lock
);
2123 idr_replace(&_minor_idr
, MINOR_ALLOCED
, MINOR(disk_devt(dm_disk(md
))));
2124 set_bit(DMF_FREEING
, &md
->flags
);
2125 spin_unlock(&_minor_lock
);
2127 blk_set_queue_dying(q
);
2129 if (dm_request_based(md
) && md
->kworker_task
)
2130 kthread_flush_worker(&md
->kworker
);
2133 * Take suspend_lock so that presuspend and postsuspend methods
2134 * do not race with internal suspend.
2136 mutex_lock(&md
->suspend_lock
);
2137 map
= dm_get_live_table(md
, &srcu_idx
);
2138 if (!dm_suspended_md(md
)) {
2139 dm_table_presuspend_targets(map
);
2140 dm_table_postsuspend_targets(map
);
2142 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2143 dm_put_live_table(md
, srcu_idx
);
2144 mutex_unlock(&md
->suspend_lock
);
2147 * Rare, but there may be I/O requests still going to complete,
2148 * for example. Wait for all references to disappear.
2149 * No one should increment the reference count of the mapped_device,
2150 * after the mapped_device state becomes DMF_FREEING.
2153 while (atomic_read(&md
->holders
))
2155 else if (atomic_read(&md
->holders
))
2156 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2157 dm_device_name(md
), atomic_read(&md
->holders
));
2160 dm_table_destroy(__unbind(md
));
2164 void dm_destroy(struct mapped_device
*md
)
2166 __dm_destroy(md
, true);
2169 void dm_destroy_immediate(struct mapped_device
*md
)
2171 __dm_destroy(md
, false);
2174 void dm_put(struct mapped_device
*md
)
2176 atomic_dec(&md
->holders
);
2178 EXPORT_SYMBOL_GPL(dm_put
);
2180 static int dm_wait_for_completion(struct mapped_device
*md
, long task_state
)
2186 prepare_to_wait(&md
->wait
, &wait
, task_state
);
2188 if (!md_in_flight(md
))
2191 if (signal_pending_state(task_state
, current
)) {
2198 finish_wait(&md
->wait
, &wait
);
2204 * Process the deferred bios
2206 static void dm_wq_work(struct work_struct
*work
)
2208 struct mapped_device
*md
= container_of(work
, struct mapped_device
,
2212 struct dm_table
*map
;
2214 map
= dm_get_live_table(md
, &srcu_idx
);
2216 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
)) {
2217 spin_lock_irq(&md
->deferred_lock
);
2218 c
= bio_list_pop(&md
->deferred
);
2219 spin_unlock_irq(&md
->deferred_lock
);
2224 if (dm_request_based(md
))
2225 generic_make_request(c
);
2227 __split_and_process_bio(md
, map
, c
);
2230 dm_put_live_table(md
, srcu_idx
);
2233 static void dm_queue_flush(struct mapped_device
*md
)
2235 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2236 smp_mb__after_atomic();
2237 queue_work(md
->wq
, &md
->work
);
2241 * Swap in a new table, returning the old one for the caller to destroy.
2243 struct dm_table
*dm_swap_table(struct mapped_device
*md
, struct dm_table
*table
)
2245 struct dm_table
*live_map
= NULL
, *map
= ERR_PTR(-EINVAL
);
2246 struct queue_limits limits
;
2249 mutex_lock(&md
->suspend_lock
);
2251 /* device must be suspended */
2252 if (!dm_suspended_md(md
))
2256 * If the new table has no data devices, retain the existing limits.
2257 * This helps multipath with queue_if_no_path if all paths disappear,
2258 * then new I/O is queued based on these limits, and then some paths
2261 if (dm_table_has_no_data_devices(table
)) {
2262 live_map
= dm_get_live_table_fast(md
);
2264 limits
= md
->queue
->limits
;
2265 dm_put_live_table_fast(md
);
2269 r
= dm_calculate_queue_limits(table
, &limits
);
2276 map
= __bind(md
, table
, &limits
);
2277 dm_issue_global_event();
2280 mutex_unlock(&md
->suspend_lock
);
2285 * Functions to lock and unlock any filesystem running on the
2288 static int lock_fs(struct mapped_device
*md
)
2292 WARN_ON(md
->frozen_sb
);
2294 md
->frozen_sb
= freeze_bdev(md
->bdev
);
2295 if (IS_ERR(md
->frozen_sb
)) {
2296 r
= PTR_ERR(md
->frozen_sb
);
2297 md
->frozen_sb
= NULL
;
2301 set_bit(DMF_FROZEN
, &md
->flags
);
2306 static void unlock_fs(struct mapped_device
*md
)
2308 if (!test_bit(DMF_FROZEN
, &md
->flags
))
2311 thaw_bdev(md
->bdev
, md
->frozen_sb
);
2312 md
->frozen_sb
= NULL
;
2313 clear_bit(DMF_FROZEN
, &md
->flags
);
2317 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2318 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2319 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2321 * If __dm_suspend returns 0, the device is completely quiescent
2322 * now. There is no request-processing activity. All new requests
2323 * are being added to md->deferred list.
2325 static int __dm_suspend(struct mapped_device
*md
, struct dm_table
*map
,
2326 unsigned suspend_flags
, long task_state
,
2327 int dmf_suspended_flag
)
2329 bool do_lockfs
= suspend_flags
& DM_SUSPEND_LOCKFS_FLAG
;
2330 bool noflush
= suspend_flags
& DM_SUSPEND_NOFLUSH_FLAG
;
2333 lockdep_assert_held(&md
->suspend_lock
);
2336 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2337 * This flag is cleared before dm_suspend returns.
2340 set_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2342 pr_debug("%s: suspending with flush\n", dm_device_name(md
));
2345 * This gets reverted if there's an error later and the targets
2346 * provide the .presuspend_undo hook.
2348 dm_table_presuspend_targets(map
);
2351 * Flush I/O to the device.
2352 * Any I/O submitted after lock_fs() may not be flushed.
2353 * noflush takes precedence over do_lockfs.
2354 * (lock_fs() flushes I/Os and waits for them to complete.)
2356 if (!noflush
&& do_lockfs
) {
2359 dm_table_presuspend_undo_targets(map
);
2365 * Here we must make sure that no processes are submitting requests
2366 * to target drivers i.e. no one may be executing
2367 * __split_and_process_bio. This is called from dm_request and
2370 * To get all processes out of __split_and_process_bio in dm_request,
2371 * we take the write lock. To prevent any process from reentering
2372 * __split_and_process_bio from dm_request and quiesce the thread
2373 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2374 * flush_workqueue(md->wq).
2376 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2378 synchronize_srcu(&md
->io_barrier
);
2381 * Stop md->queue before flushing md->wq in case request-based
2382 * dm defers requests to md->wq from md->queue.
2384 if (dm_request_based(md
)) {
2385 dm_stop_queue(md
->queue
);
2386 if (md
->kworker_task
)
2387 kthread_flush_worker(&md
->kworker
);
2390 flush_workqueue(md
->wq
);
2393 * At this point no more requests are entering target request routines.
2394 * We call dm_wait_for_completion to wait for all existing requests
2397 r
= dm_wait_for_completion(md
, task_state
);
2399 set_bit(dmf_suspended_flag
, &md
->flags
);
2402 clear_bit(DMF_NOFLUSH_SUSPENDING
, &md
->flags
);
2404 synchronize_srcu(&md
->io_barrier
);
2406 /* were we interrupted ? */
2410 if (dm_request_based(md
))
2411 dm_start_queue(md
->queue
);
2414 dm_table_presuspend_undo_targets(map
);
2415 /* pushback list is already flushed, so skip flush */
2422 * We need to be able to change a mapping table under a mounted
2423 * filesystem. For example we might want to move some data in
2424 * the background. Before the table can be swapped with
2425 * dm_bind_table, dm_suspend must be called to flush any in
2426 * flight bios and ensure that any further io gets deferred.
2429 * Suspend mechanism in request-based dm.
2431 * 1. Flush all I/Os by lock_fs() if needed.
2432 * 2. Stop dispatching any I/O by stopping the request_queue.
2433 * 3. Wait for all in-flight I/Os to be completed or requeued.
2435 * To abort suspend, start the request_queue.
2437 int dm_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2439 struct dm_table
*map
= NULL
;
2443 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2445 if (dm_suspended_md(md
)) {
2450 if (dm_suspended_internally_md(md
)) {
2451 /* already internally suspended, wait for internal resume */
2452 mutex_unlock(&md
->suspend_lock
);
2453 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2459 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2461 r
= __dm_suspend(md
, map
, suspend_flags
, TASK_INTERRUPTIBLE
, DMF_SUSPENDED
);
2465 dm_table_postsuspend_targets(map
);
2468 mutex_unlock(&md
->suspend_lock
);
2472 static int __dm_resume(struct mapped_device
*md
, struct dm_table
*map
)
2475 int r
= dm_table_resume_targets(map
);
2483 * Flushing deferred I/Os must be done after targets are resumed
2484 * so that mapping of targets can work correctly.
2485 * Request-based dm is queueing the deferred I/Os in its request_queue.
2487 if (dm_request_based(md
))
2488 dm_start_queue(md
->queue
);
2495 int dm_resume(struct mapped_device
*md
)
2498 struct dm_table
*map
= NULL
;
2502 mutex_lock_nested(&md
->suspend_lock
, SINGLE_DEPTH_NESTING
);
2504 if (!dm_suspended_md(md
))
2507 if (dm_suspended_internally_md(md
)) {
2508 /* already internally suspended, wait for internal resume */
2509 mutex_unlock(&md
->suspend_lock
);
2510 r
= wait_on_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
, TASK_INTERRUPTIBLE
);
2516 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2517 if (!map
|| !dm_table_get_size(map
))
2520 r
= __dm_resume(md
, map
);
2524 clear_bit(DMF_SUSPENDED
, &md
->flags
);
2526 mutex_unlock(&md
->suspend_lock
);
2532 * Internal suspend/resume works like userspace-driven suspend. It waits
2533 * until all bios finish and prevents issuing new bios to the target drivers.
2534 * It may be used only from the kernel.
2537 static void __dm_internal_suspend(struct mapped_device
*md
, unsigned suspend_flags
)
2539 struct dm_table
*map
= NULL
;
2541 lockdep_assert_held(&md
->suspend_lock
);
2543 if (md
->internal_suspend_count
++)
2544 return; /* nested internal suspend */
2546 if (dm_suspended_md(md
)) {
2547 set_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2548 return; /* nest suspend */
2551 map
= rcu_dereference_protected(md
->map
, lockdep_is_held(&md
->suspend_lock
));
2554 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2555 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2556 * would require changing .presuspend to return an error -- avoid this
2557 * until there is a need for more elaborate variants of internal suspend.
2559 (void) __dm_suspend(md
, map
, suspend_flags
, TASK_UNINTERRUPTIBLE
,
2560 DMF_SUSPENDED_INTERNALLY
);
2562 dm_table_postsuspend_targets(map
);
2565 static void __dm_internal_resume(struct mapped_device
*md
)
2567 BUG_ON(!md
->internal_suspend_count
);
2569 if (--md
->internal_suspend_count
)
2570 return; /* resume from nested internal suspend */
2572 if (dm_suspended_md(md
))
2573 goto done
; /* resume from nested suspend */
2576 * NOTE: existing callers don't need to call dm_table_resume_targets
2577 * (which may fail -- so best to avoid it for now by passing NULL map)
2579 (void) __dm_resume(md
, NULL
);
2582 clear_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2583 smp_mb__after_atomic();
2584 wake_up_bit(&md
->flags
, DMF_SUSPENDED_INTERNALLY
);
2587 void dm_internal_suspend_noflush(struct mapped_device
*md
)
2589 mutex_lock(&md
->suspend_lock
);
2590 __dm_internal_suspend(md
, DM_SUSPEND_NOFLUSH_FLAG
);
2591 mutex_unlock(&md
->suspend_lock
);
2593 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush
);
2595 void dm_internal_resume(struct mapped_device
*md
)
2597 mutex_lock(&md
->suspend_lock
);
2598 __dm_internal_resume(md
);
2599 mutex_unlock(&md
->suspend_lock
);
2601 EXPORT_SYMBOL_GPL(dm_internal_resume
);
2604 * Fast variants of internal suspend/resume hold md->suspend_lock,
2605 * which prevents interaction with userspace-driven suspend.
2608 void dm_internal_suspend_fast(struct mapped_device
*md
)
2610 mutex_lock(&md
->suspend_lock
);
2611 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2614 set_bit(DMF_BLOCK_IO_FOR_SUSPEND
, &md
->flags
);
2615 synchronize_srcu(&md
->io_barrier
);
2616 flush_workqueue(md
->wq
);
2617 dm_wait_for_completion(md
, TASK_UNINTERRUPTIBLE
);
2619 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast
);
2621 void dm_internal_resume_fast(struct mapped_device
*md
)
2623 if (dm_suspended_md(md
) || dm_suspended_internally_md(md
))
2629 mutex_unlock(&md
->suspend_lock
);
2631 EXPORT_SYMBOL_GPL(dm_internal_resume_fast
);
2633 /*-----------------------------------------------------------------
2634 * Event notification.
2635 *---------------------------------------------------------------*/
2636 int dm_kobject_uevent(struct mapped_device
*md
, enum kobject_action action
,
2639 char udev_cookie
[DM_COOKIE_LENGTH
];
2640 char *envp
[] = { udev_cookie
, NULL
};
2643 return kobject_uevent(&disk_to_dev(md
->disk
)->kobj
, action
);
2645 snprintf(udev_cookie
, DM_COOKIE_LENGTH
, "%s=%u",
2646 DM_COOKIE_ENV_VAR_NAME
, cookie
);
2647 return kobject_uevent_env(&disk_to_dev(md
->disk
)->kobj
,
2652 uint32_t dm_next_uevent_seq(struct mapped_device
*md
)
2654 return atomic_add_return(1, &md
->uevent_seq
);
2657 uint32_t dm_get_event_nr(struct mapped_device
*md
)
2659 return atomic_read(&md
->event_nr
);
2662 int dm_wait_event(struct mapped_device
*md
, int event_nr
)
2664 return wait_event_interruptible(md
->eventq
,
2665 (event_nr
!= atomic_read(&md
->event_nr
)));
2668 void dm_uevent_add(struct mapped_device
*md
, struct list_head
*elist
)
2670 unsigned long flags
;
2672 spin_lock_irqsave(&md
->uevent_lock
, flags
);
2673 list_add(elist
, &md
->uevent_list
);
2674 spin_unlock_irqrestore(&md
->uevent_lock
, flags
);
2678 * The gendisk is only valid as long as you have a reference
2681 struct gendisk
*dm_disk(struct mapped_device
*md
)
2685 EXPORT_SYMBOL_GPL(dm_disk
);
2687 struct kobject
*dm_kobject(struct mapped_device
*md
)
2689 return &md
->kobj_holder
.kobj
;
2692 struct mapped_device
*dm_get_from_kobject(struct kobject
*kobj
)
2694 struct mapped_device
*md
;
2696 md
= container_of(kobj
, struct mapped_device
, kobj_holder
.kobj
);
2698 spin_lock(&_minor_lock
);
2699 if (test_bit(DMF_FREEING
, &md
->flags
) || dm_deleting_md(md
)) {
2705 spin_unlock(&_minor_lock
);
2710 int dm_suspended_md(struct mapped_device
*md
)
2712 return test_bit(DMF_SUSPENDED
, &md
->flags
);
2715 int dm_suspended_internally_md(struct mapped_device
*md
)
2717 return test_bit(DMF_SUSPENDED_INTERNALLY
, &md
->flags
);
2720 int dm_test_deferred_remove_flag(struct mapped_device
*md
)
2722 return test_bit(DMF_DEFERRED_REMOVE
, &md
->flags
);
2725 int dm_suspended(struct dm_target
*ti
)
2727 return dm_suspended_md(dm_table_get_md(ti
->table
));
2729 EXPORT_SYMBOL_GPL(dm_suspended
);
2731 int dm_noflush_suspending(struct dm_target
*ti
)
2733 return __noflush_suspending(dm_table_get_md(ti
->table
));
2735 EXPORT_SYMBOL_GPL(dm_noflush_suspending
);
2737 struct dm_md_mempools
*dm_alloc_md_mempools(struct mapped_device
*md
, enum dm_queue_mode type
,
2738 unsigned integrity
, unsigned per_io_data_size
)
2740 struct dm_md_mempools
*pools
= kzalloc_node(sizeof(*pools
), GFP_KERNEL
, md
->numa_node_id
);
2741 unsigned int pool_size
= 0;
2742 unsigned int front_pad
;
2748 case DM_TYPE_BIO_BASED
:
2749 case DM_TYPE_DAX_BIO_BASED
:
2750 pool_size
= dm_get_reserved_bio_based_ios();
2751 front_pad
= roundup(per_io_data_size
, __alignof__(struct dm_target_io
)) + offsetof(struct dm_target_io
, clone
);
2753 pools
->io_pool
= mempool_create_slab_pool(pool_size
, _io_cache
);
2754 if (!pools
->io_pool
)
2757 case DM_TYPE_REQUEST_BASED
:
2758 case DM_TYPE_MQ_REQUEST_BASED
:
2759 pool_size
= dm_get_reserved_rq_based_ios();
2760 front_pad
= offsetof(struct dm_rq_clone_bio_info
, clone
);
2761 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2767 pools
->bs
= bioset_create(pool_size
, front_pad
, BIOSET_NEED_RESCUER
);
2771 if (integrity
&& bioset_integrity_create(pools
->bs
, pool_size
))
2777 dm_free_md_mempools(pools
);
2782 void dm_free_md_mempools(struct dm_md_mempools
*pools
)
2787 mempool_destroy(pools
->io_pool
);
2790 bioset_free(pools
->bs
);
2802 static int dm_call_pr(struct block_device
*bdev
, iterate_devices_callout_fn fn
,
2805 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2806 struct dm_table
*table
;
2807 struct dm_target
*ti
;
2808 int ret
= -ENOTTY
, srcu_idx
;
2810 table
= dm_get_live_table(md
, &srcu_idx
);
2811 if (!table
|| !dm_table_get_size(table
))
2814 /* We only support devices that have a single target */
2815 if (dm_table_get_num_targets(table
) != 1)
2817 ti
= dm_table_get_target(table
, 0);
2820 if (!ti
->type
->iterate_devices
)
2823 ret
= ti
->type
->iterate_devices(ti
, fn
, data
);
2825 dm_put_live_table(md
, srcu_idx
);
2830 * For register / unregister we need to manually call out to every path.
2832 static int __dm_pr_register(struct dm_target
*ti
, struct dm_dev
*dev
,
2833 sector_t start
, sector_t len
, void *data
)
2835 struct dm_pr
*pr
= data
;
2836 const struct pr_ops
*ops
= dev
->bdev
->bd_disk
->fops
->pr_ops
;
2838 if (!ops
|| !ops
->pr_register
)
2840 return ops
->pr_register(dev
->bdev
, pr
->old_key
, pr
->new_key
, pr
->flags
);
2843 static int dm_pr_register(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2854 ret
= dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2855 if (ret
&& new_key
) {
2856 /* unregister all paths if we failed to register any path */
2857 pr
.old_key
= new_key
;
2860 pr
.fail_early
= false;
2861 dm_call_pr(bdev
, __dm_pr_register
, &pr
);
2867 static int dm_pr_reserve(struct block_device
*bdev
, u64 key
, enum pr_type type
,
2870 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2871 const struct pr_ops
*ops
;
2875 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2879 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2880 if (ops
&& ops
->pr_reserve
)
2881 r
= ops
->pr_reserve(bdev
, key
, type
, flags
);
2889 static int dm_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
2891 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2892 const struct pr_ops
*ops
;
2896 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2900 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2901 if (ops
&& ops
->pr_release
)
2902 r
= ops
->pr_release(bdev
, key
, type
);
2910 static int dm_pr_preempt(struct block_device
*bdev
, u64 old_key
, u64 new_key
,
2911 enum pr_type type
, bool abort
)
2913 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2914 const struct pr_ops
*ops
;
2918 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2922 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2923 if (ops
&& ops
->pr_preempt
)
2924 r
= ops
->pr_preempt(bdev
, old_key
, new_key
, type
, abort
);
2932 static int dm_pr_clear(struct block_device
*bdev
, u64 key
)
2934 struct mapped_device
*md
= bdev
->bd_disk
->private_data
;
2935 const struct pr_ops
*ops
;
2939 r
= dm_grab_bdev_for_ioctl(md
, &bdev
, &mode
);
2943 ops
= bdev
->bd_disk
->fops
->pr_ops
;
2944 if (ops
&& ops
->pr_clear
)
2945 r
= ops
->pr_clear(bdev
, key
);
2953 static const struct pr_ops dm_pr_ops
= {
2954 .pr_register
= dm_pr_register
,
2955 .pr_reserve
= dm_pr_reserve
,
2956 .pr_release
= dm_pr_release
,
2957 .pr_preempt
= dm_pr_preempt
,
2958 .pr_clear
= dm_pr_clear
,
2961 static const struct block_device_operations dm_blk_dops
= {
2962 .open
= dm_blk_open
,
2963 .release
= dm_blk_close
,
2964 .ioctl
= dm_blk_ioctl
,
2965 .getgeo
= dm_blk_getgeo
,
2966 .pr_ops
= &dm_pr_ops
,
2967 .owner
= THIS_MODULE
2970 static const struct dax_operations dm_dax_ops
= {
2971 .direct_access
= dm_dax_direct_access
,
2972 .copy_from_iter
= dm_dax_copy_from_iter
,
2978 module_init(dm_init
);
2979 module_exit(dm_exit
);
2981 module_param(major
, uint
, 0);
2982 MODULE_PARM_DESC(major
, "The major number of the device mapper");
2984 module_param(reserved_bio_based_ios
, uint
, S_IRUGO
| S_IWUSR
);
2985 MODULE_PARM_DESC(reserved_bio_based_ios
, "Reserved IOs in bio-based mempools");
2987 module_param(dm_numa_node
, int, S_IRUGO
| S_IWUSR
);
2988 MODULE_PARM_DESC(dm_numa_node
, "NUMA node for DM device memory allocations");
2990 MODULE_DESCRIPTION(DM_NAME
" driver");
2991 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2992 MODULE_LICENSE("GPL");